As part of the Data Innovation Showcase 2024, this workshop and the relevant code aim to explore "unconditional image generation". It is the task of generating an image without being conditioned on any additional information, such as labels, text or other images that can be used as prompts. Foe example, a model trained on understanding patterns in flowers is able to create new flowers that did not exist in its training dataset.
The code in 2024DataInnovationShowcase_ImageGen.ipynb has the following learning goals:
- Explain how latent diffusion models generate images
- Show how to implement unconditional image generation using PyTorch
- Explore the use of HuggingFace for model development and inference